OpenPlan — Waze for AI agents

How it works

Three calls. One cycle.

OpenPlan exposes three MCP tools. Each does exactly one thing. Together they form a complete project lifecycle.

plan()

Decompose a goal into costed phases with estimates from the global calibration pool. Returns a route with phases, confidence intervals, evidence (hazards), personal bias, and archived routes.

plan(
  goal="Build a landing page",
  context="Astro + Tailwind",
  project="marketing-site"
)

checkpoint()

Record phase completion with actual cost. Returns deviation, hazards, and next pending phase. Correct previous costs with correct=. Check status anytime with no args.

checkpoint(
  phase="Scaffold",
  actual_cost=2100,
  route_id="uuid"
)

Correct a phase cost:

checkpoint(
  phase="Scaffold",
  correct=1950,
  route_id="uuid"
)

review()

Full retrospective with deviation analysis, accuracy metrics, cost/path learning, and diagnostics. Calibration data feeds back to the global Mesh.

review(
  route_id="uuid"
)

Why OpenPlan

Designed for the way agents work.

Every principle serves one goal: make agents better at estimating, tracking, and learning from software projects.

Local-first

All data in local SQLite. Fully offline. Auto-created. Zero setup.

SQL over ML

Cost and path learning are SQL aggregates. No models, no vectors, no training data.

MIT licensed

The MCP server is free forever. Zero gating. No feature paywalls.

Mesh learning

Anonymous calibration events sync to a global pool. Pool poisoning defenses (MAD filter, Bayesian shrinkage, per-key rate limiting) keep estimates reliable.

Graceful degradation

No errors for the agent. Sync failures are silent. Internal validation returns structured responses.

Agent loop

Plan. Execute. Checkpoint. Review.

One call to plan. One call per phase to checkpoint. One call to review. That's the whole loop — and every loop makes the next one more accurate.

Architecture

Local server. Global mesh.

Every agent contributes. Every agent benefits. The more projects you ship, the sharper the estimates get.

PRICING

Free for everyone. Better with Pro.

The server is free and always will be. Pro gives you a personal view into the global calibration data — estimates tuned to how you actually build.

Free $0

3 MCP tools: plan, checkpoint, review
Local SQLite (unlimited, offline)
Global calibration pool (100 pulls/day)
Pool-only baselines — same estimates for everyone
Mesh sync toggle via CLI
Anonymous — no account needed

Install now ↗

Pro $9 /month

Everything in Free
Personalized estimates — learns from your actual history
Unlimited baseline pulls, priority queue
Export your data (JSON, CSV, Markdown)
Roaming — authenticate once, Pro follows you

Waze for AI agents.